A reading of this essay is featured on the Metaculus Journal Podcast here.

In a recent article about solar power, Tom Chivers described how growth in solar power has outpaced many forecasts, as well as the challenges involved in accurately forecasting trends in solar power. Making accurate forecasts about the future of solar is important for understanding what future CO₂ emissions may look like and what our chances of mitigating the effects of climate change are. In this essay I attempt to estimate the bounds of solar and wind growth.

Pieces of the puzzle

The first stage of putting together an energy forecast is to understand the relevant background information. What the different units mean, how different energy types can be compared directly, and what the previous trends look like are all essential if we want to understand the future of electricity generation.

Power and energy

When electricity generation is talked about it can be easy to confuse power and energy. Power is a measure of the rate at which work is being done. Power is commonly measured in watts when discussing electricity generation. Energy is how much work has been done, or the amount of power exerted over time. It is commonly measured in watt-hours, as it is the amount of power (watts) multiplied by the amount of time that power was exerted (hours). We can imagine power as the rate of water flowing out of a hose, while energy is the amount of water in the bucket that the hose is filling after a given period of time.

These are important concepts when talking about electricity generation. When a new power plant is installed it’s usually described in terms of either its power or its capacity. But discussions about what a power plant produces are in terms of energy–such as descriptions of how much energy the plant generated in a given year.

Units

The base units for power and energy when describing electricity are typically watts and watt-hours, respectively. But on the scale of power plants or the total global power generating capacity, it’s useful to use metric prefixes. The table below shows the prefix conversions.

Table 1: Unit Conversions

Units of energy are similar. For example, 1,000 watt-hours = 1 kilowatt-hour. The table below shows energy units with examples of energy measurements of a similar magnitude in order to provide a sense of scale.

Table 2: Energy Scale

Capacity factors

The next concept to grasp is that power is not constant. Usually the power or capacity figure provided for a power plant describes the peak amount of power that plant can produce. That doesn’t mean it’s operating at that power level at all times, or even most of the time. The ratio of the amount of energy actually produced in a given time period to the amount of energy possible under ideal conditions is called the capacity factor. An easy way to understand this concept is that for a solar power plant, the plant is only producing power when the sun is shining. If the solar plant can produce full power when the sun is shining and no power when the sun is not shining, and the sun shone for exactly 12 hours in a certain day, the plant’s capacity factor would be 50% for that day.

In the real world things get messier than that. All electricity sources have a capacity factor less than 100%. The power market determines the amount of energy supplied by each plant, based on rate and other factors. A more expensive source of energy may be ramped down on days when demand is low and ramped up on days when demand is high. Other factors such as maintenance or equipment failures contribute to how high or low a power plant’s capacity factors are.

The capacity factor for a plant over the course of a year is calculated as follows:

So for a power plant with 300 megawatts of power generating capacity that produces 1,300 gigawatt-hours of energy in the course of a year, the capacity factor would be:

Tables of annual capacity factors for different energy sources and different countries are provided in this Wikipedia article. 

Previous trends

Over the last 40 years global electricity generation has shown a fairly stable rise, with an average increase of around 460 terawatt-hours of energy produced per year. Note that we’re only looking at electricity generation, as opposed to primary energy. Primary energy would include fuels such as oil and gas used to power vehicles and ships, as well as natural gas used for heating buildings.

Figure 1: Annual Electricity Generation Over Time

Coal has been the dominant source of electricity in recent decades, though in the last few years it has begun to decline as a share of total energy produced, while renewable energy sources such as wind and solar have begun to see a rapid increase.

Figure 2: Share of Electricity Production by Source Over Time

The change in capacity in recent years has begun to be dominated by renewable energy sources, with renewables rising from around 60% of new capacity brought online in 2015 to around 80% of new capacity brought online in 2020.

Figure 3: IRENA Renewable Share of Annual Power Capacity Expansion

Defining the boundaries

The above provides enough information to roughly define some boundaries: The BP Statistical Review of World Energy provides annual data for electricity generated by source, including a spreadsheet summarizing all of the data. I’ve looked at the trends in closer detail and have made the following core assumptions:

• The growth in electricity generation has been fairly level. I think we can fairly confidently assume steady growth of around 550 TWh of additional electricity generated each year.
• The changes in electricity generated by different sources each year must sum to provide this additional 550 TWh of additional electricity each year.
• We can use previous fossil fuel trends to project future rates at which fossil fuels might decline. That is, the rate that coal energy increased in the past could give a rough estimate of how quickly coal plants were brought online, and this itself might suggest a rate at which these plants would normally be retired in the future.

The above assumptions provide some constraints in the growth of renewables, and using these assumptions I’ve constructed a rough model of what the future of solar and wind power may look like.

Figure 4: Change in Total Global Electricity Generation by Year (TWh)

The mild scenario

I see this scenario as approximately a lower bound to renewable energy growth. Under this scenario I attempt to project forward from previous trends in the annual change of fossil fuels, nuclear, and hydro while assuming that global electricity demand increases annually by 550 TWh. Then I estimate what that leaves behind for wind and solar. Figures 5 and 6 show what I’ve assumed for annual generation from fossil fuels over time. (All of the data I’m using came from the BP Statistical Review of World Energy and can be seen in the Google Sheet I created here.) I assume natural gas continues to grow at an annual rate of 180 TWh and coal continues to decline, losing 300 TWh of energy generation per year from 2025 on. Under this scenario fossil fuels would decline from 62% of all electricity generated in 2020 to 48% in 2030. The amount of electricity generated by coal would decline by about 25% over that period.

Figure 5: Annual Change in Generation (TWh) - Mild - Fossil Fuels

Figure 6: Annual Generation (TWh) - Mild - Fossil Fuels

I’ve also assumed the trends in nuclear and hydro remain steady, with each of these showing slow growth. After defining these growth trends, solar and wind growth is just the remainder needed to hit the assumed 550 TWh of electricity generation growth each year. I made a simple assumption that solar would continue to grow faster than wind, with 60% of new renewables generation going to solar and the other 40% going to wind. (This proportion is somewhat arbitrary, and arguments could be made for a different distribution). As Figures 7 and 8 show, this results in a continued steep rise in solar and also in wind, though the annual change levels off around 2025 when coal reaches the steady decline I assumed of 300 TWh a year.

Figure 7: Annual Change in Generation (TWh) - Mild - Low Carbon Fuels

Figure 8: Annual Generation (TWh) - Mild - Low Carbon Fuels

The aggressive scenario

I see this scenario as an estimate of the upper bound to solar growth. In this scenario I assume more aggressive declines to fossil fuels, allowing renewables to have larger growth. I assume that instead of continuing to expand, growth in natural gas begins to slow, and overall natural gas generation begins to decline after a few years. I assume coal declines more aggressively, decreasing by 600 TWh of annual electricity generation each year from 2025 on. For nuclear and hydro I assume the same modest expansion of generation based on previous trends. I also assume the same 60% to 40% split used in the mild scenario for new renewable generation between solar and wind, respectively. These assumptions allow solar and wind to grow more significantly than under the mild scenario.

Under this scenario the share of electricity generated by fossil fuels would decline from 62% in 2020 to 35% in 2030. In 2040, fossil fuels would account for just 8% of electricity generated. Coal capacity would decline by almost 50% by 2030, and coal electricity generation would be essentially zero by 2038.

Figure 9: Annual Change in Generation (TWh) - Aggressive - Fossil Fuels

Figure 10: Annual Generation (TWh) - Aggressive - Fossil Fuels

As shown in Figures 11 and 12, solar power generation would grow by around 730 TWh in 2030. In the mild scenario this number would have been 342 TWh. Solar and wind become the clearly dominant sources of electricity generation under the aggressive scenario as well. Solar alone would generate just under 40% of all electricity in 2040 under the aggressive scenario, while under the mild scenario solar would generate just under 20% of all electricity in 2040.

Figure 11: Annual Change in Generation (TWh) - Aggressive - Low Carbon Fuels

Figure 12: Annual Generation (TWh) - Aggressive - Low Carbon Fuels

Are these assumptions reasonable?

There are benefits and drawbacks to the assumptions I’ve made. One drawback is they’re primarily trend projections. They don’t incorporate energy price projections or any other information that impacts the speed at which solar might expand. But using trend projections is also a benefit as well. It’s simpler and relies on clearly defined boundaries. We know that these trends have to be constrained by the overall growth in energy, the trend of which has been quite stable. Costs are important, but political factors are just as important. It doesn’t matter as much what the relative costs are between coal power generation and renewable power generation if regulators favor clean energy. Sometimes these regulatory decisions are reflected in the costs through clean energy subsidies, carbon taxes, and emissions regulation, but political decisions like coal phase-outs may not show up as clearly in energy prices.

In support of the common assumptions to both models, see Figure 13, which shows the annual change in electricity generation using a five-year rolling average. Total electricity generation, nuclear generation, and hydro generation have remained fairly constant over the last two decades. Assuming growth of 550 TWh per year for total electricity generation, 25 TWh per year for nuclear, and 85 TWh per year for hydro for both the mild and aggressive scenarios seems reasonable based on these trends. Ideally nuclear power would play a larger role in the future, but nuclear energy faces obstacles such as political challenges and an aging reactor fleet. The trend in overall electricity demand seems unlikely to falter as well, due to growing energy demand in developing countries as well as potential contributions from vehicle electrification. (However, this effect may be small.)

Figure 13: Annual Change in Electricity by Source (5 Year Rolling Average)

The trends in fossil fuels have greater variability. As the figure below shows, gas has held fairly steady at just under 200 TWh of additional energy generation each year and oil has seen a steady decline at a fairly constant reduction of around 50 TWh per year. Coal has seen the biggest changes, dropping from around 300 TWh of energy generation added per year in the early 2000’s to nearly zero additional energy generation in recent years.

Figure 14: 5 Year Rolling Average Change in Electricity Generation (TWh)

Based on these trends, for both scenarios I assumed energy generated from oil would decline by about 55 TWh each year, eventually reaching zero electricity generation from oil in 2035. Assumptions about gas and coal differ between the two scenarios. I assumed in the mild scenario that natural gas, which produces less than half the CO₂ produced by coal per unit of electricity generated, may be seen as an acceptable means to meet the growing demand for electricity in the coming decades. Under the mild scenario I assumed it would grow by a constant 180 TWh per year, which is near the trend for gas over the last decade. In the aggressive scenario I assumed that natural gas additions would begin to trend down as the world pushed for carbon neutrality, eventually turning negative in 2025 and on. For the rate of decline, I fit a polynomial curve to the historical data and projected outward, as shown in the figure below. While the projection is somewhat arbitrary, I think it makes for a reasonable assumption of the effect of aggressive decarbonization on natural gas.

Figure 15: Annual Change in Electricity Produced by Natural Gas (TWh) - Historical and Projected Aggressive Scenario

For coal the mild and aggressive scenarios also differ significantly. Under the mild scenario I assumed a declining trend in annual change until it reaches a reduction of 300 TWh a year, after which it maintains an annual loss of 300 TWh a year. I consider this to be the mild case because it’s a rough approximation of the rate of increase in the early 2000s, based on the assumption that coal plants reaching the end of their lifetime might be taken offline and replaced with cleaner energy sources at around the same pace coal generation was growing in the early 2000s. The aggressive case assumes the world will push for more decarbonization and will more readily avoid building new coal plants. Further, we may be more aggressive in taking older coal plants offline. This case assumes a reduction of 600 TWh of coal generation capacity from 2025 on. As previously mentioned, under this scenario electricity generation from coal would be zero by 2035, which seems reasonable as an aggressive goal. In reality, coal may always play some small role in world electricity generation even if the planet aggressively decarbonizes, but for the purposes of this estimate I’ve allowed it to reach zero.

Figure 16: Annual Change in Electricity Generated from Coal (TWh)

Lastly, we need to check that these assumptions result in reasonable trajectories for wind and solar power. As said earlier, I created projections for wind and solar by making the sum of their electricity generation equal to the amount of energy needed to satisfy the 550 TWh a year increase in total electricity demand per year. Then I assumed that 60% of the new wind and solar generation would go to solar power and 40% to wind. This assumption was primarily based on the growth in solar relative to wind in recent years.

The chart below shows what these projections would look like for the mild and aggressive cases. As shown, the aggressive case tracks pretty closely with an exponential fit of the historical trend in solar before diverging around 2025. This seems like a reasonable upper bound. It maintains its exponential growth for a few more years, after which it’s constrained by matching the pace of world energy growth and the reduction in fossil fuels. Even if the world aggressively decarbonizes, developing countries may still require new installations of coal and gas to meet their electricity needs for some time. Therefore, I think it’s unlikely that solar would continue exponentially growing for decades, as that would imply that many coal and natural gas plants would be shutting down well before they would typically be retired. The mild case sees more linear growth which eventually flattens out. I think this works well as a somewhat pessimistic lower bound where developing countries continue installing coal and gas power while developed countries are slow to decommission their fossil fuels.

Figure 17: Annual Change in Electricity Generated from Solar and Wind - Mild and Aggressive (TWh)

Wind also looks reasonable in both cases. The aggressive case looks like an upward departure from trend, while the mild case looks like a continuation of the trend. The 60% to 40% split of solar to wind installation seems to fit the trends decently enough. That said, other assumptions may be reasonable, especially since solar has seen a much steeper decline in costs than wind in recent years–though solar and wind are roughly equal in price.

Comparing to other solar projections

The next test of the mild and aggressive models is to see how they would compare to other projections of the growth in solar and wind power. Here are the sources we’ll be comparing to and the publications the data is pulled from:

• The International Energy Agency (IEA), using data published in the World Energy Outlook 2021
• The International Renewable Energy Agency (IRENA), using data published in the Global Renewables Outlook: Energy Transformation 2050
• BP, using data published in the Energy Outlook 2020 Edition

But before we can make comparisons we have to straighten out the units, because some of these sources are using power (in gigawatts) while others are using energy (in terawatt-hours).

Estimating capacity factors

To convert units so that we’re comparing apples to apples, we have to use capacity factors. I’ve chosen to compare in terms of energy generated measured in terawatt-hours. To do that I have to convert from gigawatts when those are provided. To pick capacity factors for wind and solar I used the installed capacity data from IRENA for wind and solar capacity in 2020, measured in gigawatts. I have compared that to BP’s data for electricity generated in 2020, measured in terawatt-hours. That yields capacity factors of 13.7% for solar and 24.8% for wind.

As an added check, IEA publishes data in both gigawatts and terawatt-hours, so we can figure out what they’re assuming for capacity factors. Doing so produces capacity factors of ~15.6% for solar and ~29.5% for wind. (Calculated capacity factors vary slightly depending on the scenario you use.) These capacity factors are based in 2030, and IEA actually assumes capacity factors aren’t fixed, with capacity factors improving for wind and solar over time. (Calculated capacity factors for 2040 from IEA’s scenarios would be ~17% for solar and ~32.7% for wind.) Assuming improving capacity factors makes sense, as they’ll depend on improvements in technology and other developments. But for the sake of simplicity I’ve used capacity factors of 13.7% for solar and 24.8% for wind when necessary to convert units for all time periods.

Comparing scenarios

Each group providing energy estimates for the future uses several different scenarios. They all claim not to be making predictions, but that instead they are offering different models of what may happen depending on how committed humanity ends up being to reducing or eliminating carbon emissions. All of them offer at least one scenario that roughly corresponds to either existing policies or a combination of existing policies and stated commitments, as well as offering other scenarios that correspond to more ambitious commitments. I won’t go into detail about what each scenario represents, but if you’d like to know more I have links to the relevant reports and screen captures of the scenario descriptions in the “Projections Comparison” tab of my Google Sheet.

Here are how the scenarios stack up. First, here’s the projected energy generated from solar in 2030 and 2040 for each scenario, sorted by how much energy solar would produce in 2030. In the IRENA scenarios, “PES” stands for “Planned Energy Scenario” and “TES” stands for “Transforming Energy Scenario”.

Figure 18: Solar Projections

As you can see my mild and aggressive scenarios fit well with the other estimates. My mild scenario is more ambitious than some of the other scenarios that assume we meet stated policies and goals, while my aggressive scenario falls near the more ambitious scenarios and between the BP and IEA “Net Zero” scenarios, which both achieve net zero carbon emissions by 2050. While my mild scenario seems more aggressive than the less ambitious scenarios, as shown in Figure 17 the mild scenario mostly just continues recent trends in solar and wind growth, and based on these trends it seems acceptable as a reasonable lower bound to me.¹

Figure 19: Wind Projections

Figure 19 shows the same information as Figure 18 but for wind generation. The figure shows that I assume lower growth for wind than other models assume. Notably, all of the other scenarios assume that wind continues to generate more electricity than solar each year through 2040, with the exception of the IEA Stated Policies scenario (solar produces 6,700 TWh in 2040 and wind produces 6,628 TWh) and the BP Net Zero scenario (solar produces 14,728 TWh in 2040 and wind produces 14,409 TWh). This is an interesting result to me. As I showed in Figures 8 and 12, in the mild scenario annual generation from solar passes annual generation from wind in 2028, and in the aggressive scenario that happens in 2026. The other models are more complex than mine, and mine simply assumes a 60% to 40% split between new solar and wind generation, respectively. But based on Figure 17 the result seems surprising, as the annual change in wind generation has grown at a fairly constant pace, while solar has shown more pronounced growth.

Figure 20: Wind + Solar Projections

Finally, Figure 20 shows combined wind and solar sorted by generation in 2030. The mild scenario is near the lower range of the scenarios while the aggressive is second-most ambitious.

To me these results show that my simple modeling has produced two scenarios that make for reasonable lower and upper bounds on growth in renewables. The split between solar and wind I’ve made is somewhat arbitrary and may be one of the weaker parts of these simple models, but the results it produces appear to be inside a realistic range.

Forecasting the future of renewables

Now that we’ve created two simple models and shown that they make fairly reasonable estimates of the upper and lower bound in renewables growth, we can use these bounds in our forecasting.

First, looking at the question of how much new solar PV capacity will be installed in 2030, I need to convert my estimates to gigawatts of capacity instead of terawatt-hours of energy generated. Doing so results in Figure 21.

Figure 21: Annual Change in Electricity Capacity for Solar and Wind - Mild and Aggressive (GW)

Using this estimate, solar capacity added in 2030 would range from around 285 GW in the mild scenario to around 605 GW in the aggressive scenario. Based on this, I’ve set my distribution with a 10% chance of less than 285 GW installed in 2030 and a 20% chance of more than 605 GW installed in 2030. This is significantly lower than the current median, which is around 815 GW. As a quick sanity check, if we assume a capacity factor for solar of 13.7% as previously discussed, and we look at the current crowd upper 75% forecast of 2,100 GW, converting that to TWh or annual electricity generated would imply that in 2030 electricity generated by solar would grow by 2,520 TWh. Given that annual electricity generation has grown by around 550 TWh a year, I think it’s safe to say that this upper 75% forecast is too high and that the estimate I’ve produced is not too low.

Similarly, for wind I’ve set a 10% chance of below 105 GW installed in 2030 and a 30% chance of above 225 GW. I’ve placed more uncertainty on my upper estimates because my projections are sensitive to factors I’ve made very rough estimates on, such as growth in global electricity demand and my assumed 60/40 split between new solar and wind generation. Because of those assumptions I want to avoid being overly confident in these projections.

Similarly we can forecast how much global energy will be generated by solar and wind in 2030, which tells us more about trends in renewables and progress on meeting CO₂ targets over the next decade. Referring to Figures 8 and 12, I forecast that in 2030 there’s a 10% chance that solar produces less than 3,950 TWh, a 10% chance wind produces less than 3,650 TWh, a 20% chance that solar produces more than 6,500 TWh, and a 30% chance that wind produces more than 5,350 TWh in 2030.

Discussion

I’ve demonstrated that a simple model of future electricity generation can produce estimates reasonably close to more complex models. My mild and aggressive scenarios are useful because they attempt to estimate the lower and upper bounds of the role of renewables in future electricity generation without having to deal with more complex factors like levelized cost of energy and future trajectories of individual countries. However, these benefits also present potential downsides. Modeling by IRENA, IEA, and BP take more factors into account, such as different scenarios in global electricity demand, the effects of energy prices, and effects from major players–such as India and China–in future energy demand and renewables installation. By ignoring these factors my models keep things simple, but may be missing valuable information.

Below are forecasts for how much global electricity demand there will be in 2030 and how much will be provided by different energy sources. The forecasts on these questions can be used to figure out how much electricity is likely to be generated by wind and solar in 2030. My forecasts are based on the range from my mild and aggressive models, but seeing the community forecasts may provide a better estimate of the trend over the next decade. Using the crowd forecasts on these questions I’ll be able to create a third simple model, the Metaculus model, and show what the trends from this model would look like and project those trends forward to estimate the electricity makeup in 2040.

---

¹ The IEA has commonly been criticized for underestimating the growth in solar, but it should be noted that there is some ambiguity. As shown in Figure 18, their recent estimates are not noticeably different from the other estimates shown. It’s true that some of their historical estimates were very low, but it’s also important to note that the figures showing growth leveling off are based on IEAs estimates of current policy at the time, and that IEA also provided other scenarios that were more aggressive. The head of the World Energy Outlook Power Sector Unit has defended IEA’s projections noting the different scenarios used and other factors. For more information about IEA’s projections I recommend this Vox article as well as articles from 2014 and 2015 from Energy Post. The article from 2014 shows IEA projections from their “450 Scenario” which at the time was a more aggressive scenario intended to keep CO₂ concentrations in the atmosphere below 450 ppm. But as shown in Figure 22, even this scenario estimated around 350 GW of installed solar capacity by 2020, while according to IRENA there was 714 GW of total solar capacity in 2020. So despite these caveats, even their past aggressive scenario significantly underestimated solar.

Figure 22: Chart from Energy Post Article With 2020 Lines